Oct 18, 2026 · 18 min · 22 segments
Greg Brockman just revealed the thinking of OpenAI on how best to release GPT 5. He spoke of a 'continuum of better AIs' and I will break it all down in simple terms. With evidence from multiple…
Yesterday, Greg Brockman, the president and co-founder of OpenAI, announced the company's ideas about releasing the models beyond GPT-4.
In the tweet, he made lots of points of which I found five to be particularly telling.
Cover all of them, of course, and bring in the outside evidence that reveals more.
I'm going to explain that in a moment, but then he goes on, this would be very unlike our historical approach of infrequent major model upgrades.
So what he's saying is that it's not all going to be released in one go.
He describes this as a safety opportunity.
So it's not like we're going to wake up overnight and GPT-5 is deployed.
But how would they make incrementally better AIs and what are subsequent checkpoints of a given training run? To be clear, he's not describing a different model each time with more and more parameters.
A checkpoint during a training run of GPT-5 would be a snapshot of the current value of the parameters of the model, a bit like its current understanding of the data.
And a subsequent checkpoint would be its updated parameters as it processes either more of the data or the same data more times.
Kind of like someone who rewatched a film and has a more nuanced understanding of it.
First, I want to answer those people who are thinking, isn't it already trained on all of the data on the internet? How can it get smarter? Now, I did cover this in more detail in my first GPT-5 video, but the short answer is this.
No, we're not yet running out of data.
In that video, I talked about how OpenAI may still have an order of magnitude more data to use.
That's 10 times more data still available.
And Ilya Sotskova, the chief scientist of OpenAI, put it like this, saying the data situation looks good.
Yesterday, Greg Brockman, the president and co-founder of OpenAI, announced the company's ideas about releasing the models beyond GPT-4.
In the tweet, he made lots of points of which I found five to be particularly telling.
Cover all of them, of course, and bring in the outside evidence that reveals more.
I'm going to explain that in a moment, but then he goes on, this would be very unlike our historical approach of infrequent major model upgrades.
So what he's saying is that it's not all going to be released in one go.
He describes this as a safety opportunity.
So it's not like we're going to wake up overnight and GPT-5 is deployed.
But how would they make incrementally better AIs and what are subsequent checkpoints of a given training run? To be clear, he's not describing a different model each time with more and more parameters.
A checkpoint during a training run of GPT-5 would be a snapshot of the current value of the parameters of the model, a bit like its current understanding of the data.
And a subsequent checkpoint would be its updated parameters as it processes either more of the data or the same data more times.
Kind of like someone who rewatched a film and has a more nuanced understanding of it.
First, I want to answer those people who are thinking, isn't it already trained on all of the data on the internet? How can it get smarter? Now, I did cover this in more detail in my first GPT-5 video, but the short answer is this.
No, we're not yet running out of data.
In that video, I talked about how OpenAI may still have an order of magnitude more data to use.
That's 10 times more data still available.
And Ilya Sotskova, the chief scientist of OpenAI, put it like this, saying the data situation looks good.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.